Model management

EdgeModelKit provides model download, cache, and device-tier helpers.

Download a registered model

import EdgeInference
import EdgeModelKit

guard let config = ModelConfig.find(modelID: "qwen3.5-9b-4bit") else {
    throw EdgeRuntimeError.modelNotFound("qwen3.5-9b-4bit")
}

try await HFDownloader.shared.download(config: config) { progress in
    print("Download progress:", progress)
}

Load from cache

let cachedURL = ModelCache.shared.cachedURL(for: config)

let engine = LLMEngine()
try await engine.loadLocal(directory: cachedURL)

Check cache state

if ModelCache.shared.isCached(config) {
    print("Cached at:", ModelCache.shared.cachedURL(for: config))
}

print("Model cache bytes:", ModelCache.shared.cacheSize(for: config))
print("Total cache bytes:", ModelCache.shared.totalCacheSize())

Evict models

try ModelCache.shared.evict(config)
try ModelCache.shared.evictAll()

Device tiers

ModelTierManager recommends a tier based on the current device.

import EdgeModelKit

let manager = ModelTierManager.shared
manager.refresh()

print(manager.currentTier)
print(manager.availableTiers)
print(manager.currentConfig)

Model categories

ModelCategory.detect(from:) reads config.json and selects an engine category.

let category = ModelCategory.detect(from: cachedURL)

switch category {
case .llm:
    print("Use LLMEngine")
case .vlm:
    print("Use VLMEngine")
case .tts:
    print("Use TTSEngine")
case .stt:
    print("Use STTEngine")
}

Loading strategy

Source	Use it when
Local path	You already have a model bundle on disk.
Model cache	The app downloaded a registered model.
Edge Studio export	You optimized or prepared the model in Edge Studio.
App bundle or ODR	The app ships or downloads the model through iOS distribution.